Has anyone seen the article "Floating-point DSCs yield greener control systems" in the Mar 2008 Electronic Products by Andrew Soukup? I just got my copy this afternoon. (DSC = Digital Signal Controller, e.g., the TI TMS320F28x family.) This article tries really hard to convince the reader that floating-point is simultaneously "greener" (more power efficient) and faster than fixed-point. Hrmph! First of all, he's basing the floating-point analysis on the TI TMS320F28335, which has a 6-channel DMA controller that, to the best of my knowledge, is not available in the fixed-point products. This alone can be responsible for a fair amount of performance improvement. Secondly, the column titles in his Table 1 (a comparison of fixed-point versus floating-point performances for various common algorithms like division, square root, etc.) sound suspicious: the fixed-point column title is "C28x 32-bit (Fixed)", while the floating-point column title is "F283x Optimized Assembly (Floating)*" - this implies to me they did not optimize the fixed-point version, otherwise why would they claim "optimized assembly" for one and not the other. Thirdly, the comparison is formed for 32-bit fixed-point operations and 32-bit floating-point operations. Last time I checked, a 32-bit fixed-point operation beats the resolution of 32-bit floating-point operations (effectively 24 bit mantissa) by about 8 bits. Lastly, although I didn't see this stated, I bet this processor was developed with a newer process technology like 45 nm CMOS. Dunno. Sounds like a LOT of marketing hype, and even bullshit, to me. Your opinions? To give Andrew a break, I'll bet he knew exactly what was being done and was pressured by TI to make this marketing statement. -- % Randy Yates % "Ticket to the moon, flight leaves here today %% Fuquay-Varina, NC % from Satellite 2" %%% 919-577-9882 % 'Ticket To The Moon' %%%% <yates@ieee.org> % *Time*, Electric Light Orchestra http://www.digitalsignallabs.com
What Was This Guy Smoking?
Started by ●March 18, 2008
Reply by ●March 19, 20082008-03-19
Hi Randy, you're right.. Everything that is not strictly a technical document (except some case) such a presentation or brochure is a marketing hype. You can only accept that this new family is a new product different from the others. If you feel curious about these products you can read some datasheets but.. technically each brochure is ridiculous.. it's not written for engineers. But in a big company brochures are important because they are the "showroom" for the company. Now you know that new products exist, you and everyone has read your post. If you'll have to develop a new application I'm sure you'll read more technical docs, if you have the spirit to try to develop with new devices. Maybe you'll discover that is more appropriately than others for your app. This is just what TI wants to do. And for this reason brochures are not written by "technicians". I think it's better to "read" brochures (or similar web pages) for less than half minute. Just to know that somewhat exists. Raffa Randy Yates wrote:> Has anyone seen the article "Floating-point DSCs yield greener control > systems" in the Mar 2008 Electronic Products by Andrew Soukup? I just > got my copy this afternoon. (DSC = Digital Signal Controller, e.g., the > TI TMS320F28x family.) > > This article tries really hard to convince the reader that > floating-point is simultaneously "greener" (more power efficient) and > faster than fixed-point. Hrmph! > > First of all, he's basing the floating-point analysis on the TI > TMS320F28335, which has a 6-channel DMA controller that, to the best of > my knowledge, is not available in the fixed-point products. This alone > can be responsible for a fair amount of performance improvement. > > Secondly, the column titles in his Table 1 (a comparison of fixed-point > versus floating-point performances for various common algorithms like > division, square root, etc.) sound suspicious: the fixed-point column > title is "C28x 32-bit (Fixed)", while the floating-point column title is > "F283x Optimized Assembly (Floating)*" - this implies to me they did not > optimize the fixed-point version, otherwise why would they claim > "optimized assembly" for one and not the other. > > Thirdly, the comparison is formed for 32-bit fixed-point operations and > 32-bit floating-point operations. Last time I checked, a 32-bit > fixed-point operation beats the resolution of 32-bit floating-point > operations (effectively 24 bit mantissa) by about 8 bits. > > Lastly, although I didn't see this stated, I bet this processor was > developed with a newer process technology like 45 nm CMOS. > > Dunno. Sounds like a LOT of marketing hype, and even bullshit, to > me. Your opinions? > > To give Andrew a break, I'll bet he knew exactly what was being done and > was pressured by TI to make this marketing statement.
Reply by ●March 19, 20082008-03-19
-= Mr. Raffa =- wrote:> Hi Randy, > you're right.. Everything that is not strictly a technical document > (except some case) such a presentation or brochure is a marketing hype. > You can only accept that this new family is a new product different from > the others. If you feel curious about these products you can read some > datasheets but.. technically each brochure is ridiculous.. it's not > written for engineers. > But in a big company brochures are important because they are the > "showroom" for the company. Now you know that new products exist, you > and everyone has read your post. If you'll have to develop a new > application I'm sure you'll read more technical docs, if you have the > spirit to try to develop with new devices. Maybe you'll discover that is > more appropriately than others for your app. > This is just what TI wants to do. And for this reason brochures are not > written by "technicians". > I think it's better to "read" brochures (or similar web pages) for less > than half minute. Just to know that somewhat exists. > > RaffaI think that's just more spin*. Any technical presentation, whether on slick paper or not, is either technically accurate, fraudulent, or mistaken. How do you rate a compan8 that says, "OK; I lied, but it was for your own good?" TI must have hired someone away from Big Pharma, Jerry _________________________________ * "Bullshit" is probably a better term. -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●March 19, 20082008-03-19
On Tue, 18 Mar 2008 16:28:49 -0400, Randy Yates <yates@ieee.org> wrote: (snipped by Lyons)> >This article tries really hard to convince the reader that >floating-point is simultaneously "greener" (more power efficient) and >faster than fixed-point. Hrmph!Hi Randy, maybe instead of "greener" the author should have said that floating-point "helps the children". :-) [-Rick-]
Reply by ●March 20, 20082008-03-20
Randy Yates wrote:> Has anyone seen the article "Floating-point DSCs yield greener control > systems" in the Mar 2008 Electronic Products by Andrew Soukup? I just > got my copy this afternoon. (DSC = Digital Signal Controller, e.g., the > TI TMS320F28x family.) > > This article tries really hard to convince the reader that > floating-point is simultaneously "greener" (more power efficient) and > faster than fixed-point. Hrmph! > > First of all, he's basing the floating-point analysis on the TI > TMS320F28335, which has a 6-channel DMA controller that, to the best of > my knowledge, is not available in the fixed-point products. This alone > can be responsible for a fair amount of performance improvement.As has been said in this forum a number of times, DMA controllers are dead weight for most control applications. Presumably these controllers have been added to broaden the appeal of the devices outside the control field. In any meaningful comparison of the fixed floating point control oriented devices they play no part. > [...] Regards, Steve
Reply by ●March 20, 20082008-03-20
Jerry Avins wrote:> I think that's just more spin*. Any technical presentation, whether on > slick paper or not, is either technically accurate, fraudulent, or > mistaken. How do you rate a compan8 that says, "OK; I lied, but it was > for your own good?" TI must have hired someone away from Big Pharma, > > Jerry > _________________________________ > * "Bullshit" is probably a better term.Hi Jerry mine it wasn't a way to justify TI or other companies that lies in this way.. (all companies?!?) Just explain what I see in my company and in many other companies.. I have a co-worker that worked in TI and some time ago he explains me the situation. Finally one choose to use the most appropriate component disregarding the marketing type of the company, just for the technical characteristics. You can use TI, Analog or everything you want. If your final customer wants a "green processor" in it's application because he read TI brochure, then you can paint it! We can understand what is a fake or not but.. Do you ever have contact with customers that doesn't know anything about this? It's the dark side of engineering.. Anyway, these companies have to say: "OK, I lied but it was for MY own good" ;-) Raffa
Reply by ●March 20, 20082008-03-20
On Mar 18, 8:28�pm, Randy Yates <ya...@ieee.org> wrote:> Has anyone seen the article "Floating-point DSCs yield greener control > systems" in the Mar 2008 Electronic Products by Andrew Soukup? I just > got my copy this afternoon. (DSC = Digital Signal Controller, e.g., the > TI TMS320F28x family.) > > This article tries really hard to convince the reader that > floating-point is simultaneously "greener" (more power efficient) and > faster than fixed-point. Hrmph! > > First of all, he's basing the floating-point analysis on the TI > TMS320F28335, which has a 6-channel DMA controller that, to the best of > my knowledge, is not available in the fixed-point products. This alone > can be responsible for a fair amount of performance improvement. > > Secondly, the column titles in his Table 1 (a comparison of fixed-point > versus floating-point performances for various common algorithms like > division, square root, etc.) sound suspicious: the fixed-point column > title is "C28x 32-bit (Fixed)", while the floating-point column title is > "F283x Optimized Assembly (Floating)*" - this implies to me they did not > optimize the fixed-point version, otherwise why would they claim > "optimized assembly" for one and not the other. > > Thirdly, the comparison is formed for 32-bit fixed-point operations and > 32-bit floating-point operations. Last time I checked, a 32-bit > fixed-point operation beats the resolution of 32-bit floating-point > operations (effectively 24 bit mantissa) by about 8 bits. > > Lastly, although I didn't see this stated, I bet this processor was > developed with a newer process technology like 45 nm CMOS. > > Dunno. Sounds like a LOT of marketing hype, and even bullshit, to > me. Your opinions? > > To give Andrew a break, I'll bet he knew exactly what was being done and > was pressured by TI to make this marketing statement. > -- > % �Randy Yates � � � � � � � � �% "Ticket to the moon, flight leaves here today > %% Fuquay-Varina, NC � � � � � �% �from Satellite 2" > %%% 919-577-9882 � � � � � � � �% 'Ticket To The Moon' > %%%% <ya...@ieee.org> � � � � � % *Time*, Electric Light Orchestrahttp://www.digitalsignallabs.comPerhaps it's potentially greener with respect to the amount of software development hours required!
Reply by ●March 20, 20082008-03-20
-= Mr. Raffa =- wrote:> Jerry Avins wrote: >> I think that's just more spin*. Any technical presentation, whether on >> slick paper or not, is either technically accurate, fraudulent, or >> mistaken. How do you rate a compan8 that says, "OK; I lied, but it was >> for your own good?" TI must have hired someone away from Big Pharma, >> >> Jerry >> _________________________________ >> * "Bullshit" is probably a better term. > > Hi Jerry > mine it wasn't a way to justify TI or other companies that lies in this > way.. (all companies?!?) Just explain what I see in my company and in > many other companies.. I have a co-worker that worked in TI and some > time ago he explains me the situation. Finally one choose to use the > most appropriate component disregarding the marketing type of the > company, just for the technical characteristics. You can use TI, Analog > or everything you want. If your final customer wants a "green processor" > in it's application because he read TI brochure, then you can paint it! > We can understand what is a fake or not but.. Do you ever have contact > with customers that doesn't know anything about this? It's the dark side > of engineering.. > Anyway, these companies have to say: "OK, I lied but it was for MY own > good" ;-)This kind of advertising is like the "Ask your doctor if xxxx is right for you" ads. Either doctors spend time explaining why xxxx isn't right or prescribing it anyway just to get on with things. Engineers spend time explaining to clients why xxxx isn't the best way, or use it anyway just to keep the client happy. Either way, misleading ads help only the advertiser, and then maybe not for long. I hope TI soon wises up and stops playing the Big Pharma game. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●March 20, 20082008-03-20
Steve Underwood wrote:> Randy Yates wrote: >> Has anyone seen the article "Floating-point DSCs yield greener control >> systems" in the Mar 2008 Electronic Products by Andrew Soukup? I just >> got my copy this afternoon. (DSC = Digital Signal Controller, e.g., the >> TI TMS320F28x family.) >> >> This article tries really hard to convince the reader that >> floating-point is simultaneously "greener" (more power efficient) and >> faster than fixed-point. Hrmph! >> >> First of all, he's basing the floating-point analysis on the TI >> TMS320F28335, which has a 6-channel DMA controller that, to the best of >> my knowledge, is not available in the fixed-point products. This alone >> can be responsible for a fair amount of performance improvement. > > As has been said in this forum a number of times, DMA controllers are > dead weight for most control applications. Presumably these controllers > have been added to broaden the appeal of the devices outside the control > field. In any meaningful comparison of the fixed floating point control > oriented devices they play no part. > > > [...] > > Regards, > SteveDoesn't that depend on the size of the control program and the controller's internal RAM? DMA is very useful for shuffling overlays. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●March 20, 20082008-03-20
Timothy Freeburn wrote:> On Mar 18, 8:28 pm, Randy Yates <ya...@ieee.org> wrote: >> Has anyone seen the article "Floating-point DSCs yield greener control >> systems" in the Mar 2008 Electronic Products by Andrew Soukup? I just >> got my copy this afternoon. (DSC = Digital Signal Controller, e.g., the >> TI TMS320F28x family.) >> >> This article tries really hard to convince the reader that >> floating-point is simultaneously "greener" (more power efficient) and >> faster than fixed-point. Hrmph! >> >> First of all, he's basing the floating-point analysis on the TI >> TMS320F28335, which has a 6-channel DMA controller that, to the best of >> my knowledge, is not available in the fixed-point products. This alone >> can be responsible for a fair amount of performance improvement. >> >> Secondly, the column titles in his Table 1 (a comparison of fixed-point >> versus floating-point performances for various common algorithms like >> division, square root, etc.) sound suspicious: the fixed-point column >> title is "C28x 32-bit (Fixed)", while the floating-point column title is >> "F283x Optimized Assembly (Floating)*" - this implies to me they did not >> optimize the fixed-point version, otherwise why would they claim >> "optimized assembly" for one and not the other. >> >> Thirdly, the comparison is formed for 32-bit fixed-point operations and >> 32-bit floating-point operations. Last time I checked, a 32-bit >> fixed-point operation beats the resolution of 32-bit floating-point >> operations (effectively 24 bit mantissa) by about 8 bits. >> >> Lastly, although I didn't see this stated, I bet this processor was >> developed with a newer process technology like 45 nm CMOS. >> >> Dunno. Sounds like a LOT of marketing hype, and even bullshit, to >> me. Your opinions? >> >> To give Andrew a break, I'll bet he knew exactly what was being done and >> was pressured by TI to make this marketing statement. >> -- >> % Randy Yates % "Ticket to the moon, flight leaves here today >> %% Fuquay-Varina, NC % from Satellite 2" >> %%% 919-577-9882 % 'Ticket To The Moon' >> %%%% <ya...@ieee.org> % *Time*, Electric Light Orchestrahttp://www.digitalsignallabs.com > > Perhaps it's potentially greener with respect to the amount of > software development hours required!Maybe he was referring to "long green". Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
